A three-dimensional direct simulation was carried out to examine the natural convective fluid flow and heat transfer characteristics in a long enclosure with a twisted cylinder. The effects of the Rayleigh number, dimensionless mean radius, and wavelength (or wavenumber) for a twisted cylinder were investigated. In addition to the heat transfer characteristics at cylinder and enclosure, the distributions of temperature and fluid flow are also presented by the iso-surfaces of temperature and vortical structure formed inside the enclosure. An artificial neural network was applied to predict the characteristics of natural convective flow and heat transfer by considering various simulation parameters for the elliptical and twisted cylinders. The input parameters considered in this study significantly affect on the heat transfer characteristics at cylinder and enclosure.